The EU reaches world-first agreement on AI regulations

As artificial intelligence (AI) continues to ignite debates as both a potential blessing and threat, European Union countries have taken steps to regulate the operations of AI-based software. This move, unprecedented on a global scale, culminated in an agreement known as the AI Act, announced on Friday, to regulate artificial intelligence's development and use.

"An agreement has been reached!" declared Thierry Breton, the EU Commissioner for Internal Market and Services, on Platform X. "This is a historic moment. The EU is the first continent to set clear rules for the use of artificial intelligence," he stated in a subsequent post.

Breton emphasized that this "political agreement" pertains to legislation aimed at promoting innovation in advanced technologies across Europe while simultaneously curbing potential misuse.

The European Commission presented the proposal for AI regulation in April 2021. This was intended to be the first comprehensive regulation of AI systems in Europe. The Council and the European Parliament then fine-tuned this proposal. To convert it into enforceable law, they had to develop compromise solutions and put the uniform act to a vote in the European Parliament.

According to the AFP agency, the negotiation process was heavily influenced by the emergence of OpenAI's text generator, ChatGPT. This Californian technology is capable of generating essays, poetry, and translations within seconds.

This system, like others capable of creating sounds or images, exposed both the enormous potential and significant risks of artificial intelligence. For instance, the spread of counterfeit photos on social media demonstrated the hazards associated with manipulating public opinion.

The European Parliament members insisted on incorporating the generative AI phenomenon in the current discussions, emphasizing the need for more strict supervision of this high-impact technology. Specifically, they called for greater transparency in the algorithms and extensive databases that form the core of these systems.

Before negotiations began, member states expressed some concern that extensive regulations could hinder the growth of local AI leaders like Aleph Alpha in Germany and Mistal AI in France.

AFP elaborates that the political agreement reached on Friday evening will be polished by technical work to finalize the text. "We will scrutinize the compromise reached today and will ensure in the upcoming weeks that the text upholds Europe's ability to nurture homegrown AI technologies and maintain its strategic autonomy," stated French Digital Minister, Jean-Noel Barrot.

The compromise plans to maintain a dual-speed approach, focusing on AI development and safeguards to prevent misuse. The proposal intends to determine the quality of data used in algorithm development and safeguard against copyright violations. It also mandates creators to label artificially created sounds, images, and texts as AI-derived.

The project is centered around a list of rules applied exclusively to "high-risk" systems, often used in sensitive sectors like critical infrastructure, education, human resources, law enforcement, etc. These systems will be subject to various obligations such as human oversight, technical documentation, and risk management system implementation.

The regulations also anticipate special supervision for AI systems that interact with humans. This includes requirements to inform users that they are engaging with a machine.

The document also makes rare provision for prohibitions, primarily targeting applications conflicting with European values. The excesses seen in China, like citizen assessment systems or mass surveillance, should be avoided. However, EU countries have been given some leeway concerning certain law enforcement activities, such as counter-terrorism.

The agreement proposes that European legislation be enforced through supervisory and penalty measures, including the establishment of a European AI office in the European Commission. The most severe offenses could face penalties of up to 7% of turnover, with the lower limit set at 35 million euros (approximately $39 million).